utils/mountd/cache.c

   1
   2 /*
   3  * Handle communication with knfsd internal cache
   4  *
   5  * We open /proc/net/rpc/{auth.unix.ip,nfsd.export,nfsd.fh}/channel
   6  * and listen for requests (using my_svc_run)
   7  *
   8  */
   9
  10 #ifdef HAVE_CONFIG_H
  11 #include <config.h>
  12 #endif
  13
  14 #include <sys/types.h>
  15 #include <sys/select.h>
  16 #include <sys/stat.h>
  17 #include <sys/vfs.h>
  18 #include <time.h>
  19 #include <netinet/in.h>
  20 #include <arpa/inet.h>
  21 #include <unistd.h>
  22 #include <fcntl.h>
  23 #include <errno.h>
  24 #include <ctype.h>
  25 #include <pwd.h>
  26 #include <grp.h>
  27 #include <mntent.h>
  28 #include "misc.h"
  29 #include "nfslib.h"
  30 #include "exportfs.h"
  31 #include "mountd.h"
  32 #include "xmalloc.h"
  33 #include "fsloc.h"
  34 #include "pseudoflavors.h"
  35
  36 #ifdef USE_BLKID
  37 #include "blkid/blkid.h"
  38 #endif
  39
  40 /*
  41  * Invoked by RPC service loop
  42  */
  43 void    cache_set_fds(fd_set *fdset);
  44 int     cache_process_req(fd_set *readfds);
  45
  46 enum nfsd_fsid {
  47         FSID_DEV = 0,
  48         FSID_NUM,
  49         FSID_MAJOR_MINOR,
  50         FSID_ENCODE_DEV,
  51         FSID_UUID4_INUM,
  52         FSID_UUID8,
  53         FSID_UUID16,
  54         FSID_UUID16_INUM,
  55 };
  56
  57 /*
  58  * Support routines for text-based upcalls.
  59  * Fields are separated by spaces.
  60  * Fields are either mangled to quote space tab newline slosh with slosh
  61  * or a hexified with a leading \x
  62  * Record is terminated with newline.
  63  *
  64  */
  65 static int cache_export_ent(char *domain, struct exportent *exp, char *p);
  66
  67 #define INITIAL_MANAGED_GROUPS 100
  68
  69 char *lbuf  = NULL;
  70 int lbuflen = 0;
  71 extern int use_ipaddr;
  72
  73 static void auth_unix_ip(FILE *f)
  74 {
  75         /* requests are
  76          *  class IP-ADDR
  77          * Ignore if class != "nfsd"
  78          * Otherwise find domainname and write back:
  79          *
  80          *  "nfsd" IP-ADDR expiry domainname
  81          */
  82         char *cp;
  83         char class[20];
  84         char ipaddr[INET6_ADDRSTRLEN];
  85         char *client = NULL;
  86         struct addrinfo *tmp = NULL;
  87         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
  88                 return;
  89
  90         xlog(D_CALL, "auth_unix_ip: inbuf '%s'", lbuf);
  91
  92         cp = lbuf;
  93
  94         if (qword_get(&cp, class, 20) <= 0 ||
  95             strcmp(class, "nfsd") != 0)
  96                 return;
  97
  98         if (qword_get(&cp, ipaddr, sizeof(ipaddr)) <= 0)
  99                 return;
 100
 101         tmp = host_pton(ipaddr);
 102         if (tmp == NULL)
 103                 return;
 104
 105         auth_reload();
 106
 107         /* addr is a valid, interesting address, find the domain name... */
 108         if (!use_ipaddr) {
 109                 struct addrinfo *ai = NULL;
 110
 111                 ai = client_resolve(tmp->ai_addr);
 112                 if (ai == NULL)
 113                         goto out;
 114                 client = client_compose(ai);
 115                 freeaddrinfo(ai);
 116                 if (!client)
 117                         goto out;
 118         }
 119         qword_print(f, "nfsd");
 120         qword_print(f, ipaddr);
 121         qword_printtimefrom(f, DEFAULT_TTL);
 122         if (use_ipaddr)
 123                 qword_print(f, ipaddr);
 124         else if (client)
 125                 qword_print(f, *client?client:"DEFAULT");
 126         qword_eol(f);
 127         xlog(D_CALL, "auth_unix_ip: client %p '%s'", client, client?client: "DEFAULT");
 128
 129         free(client);
 130 out:
 131         freeaddrinfo(tmp);
 132
 133 }
 134
 135 static void auth_unix_gid(FILE *f)
 136 {
 137         /* Request are
 138          *  uid
 139          * reply is
 140          *  uid expiry count list of group ids
 141          */
 142         uid_t uid;
 143         struct passwd *pw;
 144         static gid_t *groups = NULL;
 145         static int groups_len = 0;
 146         gid_t *more_groups;
 147         int ngroups;
 148         int rv, i;
 149         char *cp;
 150
 151         if (groups_len == 0) {
 152                 groups = malloc(sizeof(gid_t) * INITIAL_MANAGED_GROUPS);
 153                 if (!groups)
 154                         return;
 155
 156                 groups_len = INITIAL_MANAGED_GROUPS;
 157         }
 158
 159         ngroups = groups_len;
 160
 161         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 162                 return;
 163
 164         cp = lbuf;
 165         if (qword_get_uint(&cp, &uid) != 0)
 166                 return;
 167
 168         pw = getpwuid(uid);
 169         if (!pw)
 170                 rv = -1;
 171         else {
 172                 rv = getgrouplist(pw->pw_name, pw->pw_gid, groups, &ngroups);
 173                 if (rv == -1 && ngroups >= groups_len) {
 174                         more_groups = realloc(groups, sizeof(gid_t)*ngroups);
 175                         if (!more_groups)
 176                                 rv = -1;
 177                         else {
 178                                 groups = more_groups;
 179                                 groups_len = ngroups;
 180                                 rv = getgrouplist(pw->pw_name, pw->pw_gid,
 181                                                   groups, &ngroups);
 182                         }
 183                 }
 184         }
 185         qword_printuint(f, uid);
 186         qword_printtimefrom(f, DEFAULT_TTL);
 187         if (rv >= 0) {
 188                 qword_printuint(f, ngroups);
 189                 for (i=0; i<ngroups; i++)
 190                         qword_printuint(f, groups[i]);
 191         } else
 192                 qword_printuint(f, 0);
 193         qword_eol(f);
 194 }
 195
 196 #if USE_BLKID
 197 static const char *get_uuid_blkdev(char *path)
 198 {
 199         /* We set *safe if we know that we need the
 200          * fsid from statfs too.
 201          */
 202         static blkid_cache cache = NULL;
 203         struct stat stb;
 204         char *devname;
 205         blkid_tag_iterate iter;
 206         blkid_dev dev;
 207         const char *type;
 208         const char *val, *uuid = NULL;
 209
 210         if (cache == NULL)
 211                 blkid_get_cache(&cache, NULL);
 212
 213         if (stat(path, &stb) != 0)
 214                 return NULL;
 215         devname = blkid_devno_to_devname(stb.st_dev);
 216         if (!devname)
 217                 return NULL;
 218         dev = blkid_get_dev(cache, devname, BLKID_DEV_NORMAL);
 219         free(devname);
 220         if (!dev)
 221                 return NULL;
 222         iter = blkid_tag_iterate_begin(dev);
 223         if (!iter)
 224                 return NULL;
 225         while (blkid_tag_next(iter, &type, &val) == 0) {
 226                 if (strcmp(type, "UUID") == 0)
 227                         uuid = val;
 228                 if (strcmp(type, "TYPE") == 0 &&
 229                     strcmp(val, "btrfs") == 0) {
 230                         uuid = NULL;
 231                         break;
 232                 }
 233         }
 234         blkid_tag_iterate_end(iter);
 235         return uuid;
 236 }
 237 #else
 238 #define get_uuid_blkdev(path) (NULL)
 239 #endif
 240
 241 static int get_uuid(const char *val, size_t uuidlen, char *u)
 242 {
 243         /* extract hex digits from uuidstr and compose a uuid
 244          * of the given length (max 16), xoring bytes to make
 245          * a smaller uuid.
 246          */
 247         size_t i = 0;
 248
 249         memset(u, 0, uuidlen);
 250         for ( ; *val ; val++) {
 251                 char c = *val;
 252                 if (!isxdigit(c))
 253                         continue;
 254                 if (isalpha(c)) {
 255                         if (isupper(c))
 256                                 c = c - 'A' + 10;
 257                         else
 258                                 c = c - 'a' + 10;
 259                 } else
 260                         c = c - '0' + 0;
 261                 if ((i&1) == 0)
 262                         c <<= 4;
 263                 u[i/2] ^= c;
 264                 i++;
 265                 if (i == uuidlen*2)
 266                         i = 0;
 267         }
 268         return 1;
 269 }
 270
 271 static int uuid_by_path(char *path, int type, size_t uuidlen, char *uuid)
 272 {
 273         /* get a uuid for the filesystem found at 'path'.
 274          * There are several possible ways of generating the
 275          * uuids (types).
 276          * Type 0 is used for new filehandles, while other types
 277          * may be used to interpret old filehandle - to ensure smooth
 278          * forward migration.
 279          * We return 1 if a uuid was found (and it might be worth
 280          * trying the next type) or 0 if no more uuid types can be
 281          * extracted.
 282          */
 283
 284         /* Possible sources of uuid are
 285          * - blkid uuid
 286          * - statfs64 uuid
 287          *
 288          * On some filesystems (e.g. vfat) the statfs64 uuid is simply an
 289          * encoding of the device that the filesystem is mounted from, so
 290          * it we be very bad to use that (as device numbers change).  blkid
 291          * must be preferred.
 292          * On other filesystems (e.g. btrfs) the statfs64 uuid contains
 293          * important info that the blkid uuid cannot contain:  This happens
 294          * when multiple subvolumes are exported (they have the same
 295          * blkid uuid but different statfs64 uuids).
 296          * We rely on get_uuid_blkdev *knowing* which is which and not returning
 297          * a uuid for filesystems where the statfs64 uuid is better.
 298          *
 299          */
 300         struct statfs64 st;
 301         char fsid_val[17];
 302         const char *blkid_val;
 303         const char *val;
 304
 305         blkid_val = get_uuid_blkdev(path);
 306
 307         if (statfs64(path, &st) == 0 &&
 308             (st.f_fsid.__val[0] || st.f_fsid.__val[1]))
 309                 snprintf(fsid_val, 17, "%08x%08x",
 310                          st.f_fsid.__val[0], st.f_fsid.__val[1]);
 311         else
 312                 fsid_val[0] = 0;
 313
 314         if (blkid_val && (type--) == 0)
 315                 val = blkid_val;
 316         else if (fsid_val[0] && (type--) == 0)
 317                 val = fsid_val;
 318         else
 319                 return 0;
 320
 321         get_uuid(val, uuidlen, uuid);
 322         return 1;
 323 }
 324
 325 /* Iterate through /etc/mtab, finding mountpoints
 326  * at or below a given path
 327  */
 328 static char *next_mnt(void **v, char *p)
 329 {
 330         FILE *f;
 331         struct mntent *me;
 332         size_t l = strlen(p);
 333         if (*v == NULL) {
 334                 f = setmntent("/etc/mtab", "r");
 335                 *v = f;
 336         } else
 337                 f = *v;
 338         while ((me = getmntent(f)) != NULL &&
 339                (strncmp(me->mnt_dir, p, l) != 0 ||
 340                 me->mnt_dir[l] != '/'))
 341                 ;
 342         if (me == NULL) {
 343                 endmntent(f);
 344                 *v = NULL;
 345                 return NULL;
 346         }
 347         return me->mnt_dir;
 348 }
 349
 350 /* True iff e1 is a child of e2 and e2 has crossmnt set: */
 351 static bool subexport(struct exportent *e1, struct exportent *e2)
 352 {
 353         char *p1 = e1->e_path, *p2 = e2->e_path;
 354         size_t l2 = strlen(p2);
 355
 356         return e2->e_flags & NFSEXP_CROSSMOUNT
 357                && strncmp(p1, p2, l2) == 0
 358                && p1[l2] == '/';
 359 }
 360
 361 struct parsed_fsid {
 362         int fsidtype;
 363         /* We could use a union for this, but it would be more
 364          * complicated; why bother? */
 365         unsigned int inode;
 366         unsigned int minor;
 367         unsigned int major;
 368         unsigned int fsidnum;
 369         size_t uuidlen;
 370         char *fhuuid;
 371 };
 372
 373 static int parse_fsid(int fsidtype, int fsidlen, char *fsid,
 374                 struct parsed_fsid *parsed)
 375 {
 376         unsigned int dev;
 377         unsigned long long inode64;
 378
 379         memset(parsed, 0, sizeof(*parsed));
 380         parsed->fsidtype = fsidtype;
 381         switch(fsidtype) {
 382         case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
 383                 if (fsidlen != 8)
 384                         return -1;
 385                 memcpy(&dev, fsid, 4);
 386                 memcpy(&parsed->inode, fsid+4, 4);
 387                 parsed->major = ntohl(dev)>>16;
 388                 parsed->minor = ntohl(dev) & 0xFFFF;
 389                 break;
 390
 391         case FSID_NUM: /* 4 bytes - fsid */
 392                 if (fsidlen != 4)
 393                         return -1;
 394                 memcpy(&parsed->fsidnum, fsid, 4);
 395                 break;
 396
 397         case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode
 398                  * This format is never actually used but was
 399                  * an historical accident
 400                  */
 401                 if (fsidlen != 12)
 402                         return -1;
 403                 memcpy(&dev, fsid, 4);
 404                 parsed->major = ntohl(dev);
 405                 memcpy(&dev, fsid+4, 4);
 406                 parsed->minor = ntohl(dev);
 407                 memcpy(&parsed->inode, fsid+8, 4);
 408                 break;
 409
 410         case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
 411                 /* This is *host* endian, not net-byte-order, because
 412                  * no-one outside this host has any business interpreting it
 413                  */
 414                 if (fsidlen != 8)
 415                         return -1;
 416                 memcpy(&dev, fsid, 4);
 417                 memcpy(&parsed->inode, fsid+4, 4);
 418                 parsed->major = (dev & 0xfff00) >> 8;
 419                 parsed->minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
 420                 break;
 421
 422         case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
 423                 if (fsidlen != 8)
 424                         return -1;
 425                 memcpy(&parsed->inode, fsid, 4);
 426                 parsed->uuidlen = 4;
 427                 parsed->fhuuid = fsid+4;
 428                 break;
 429         case FSID_UUID8: /* 8 byte uuid */
 430                 if (fsidlen != 8)
 431                         return -1;
 432                 parsed->uuidlen = 8;
 433                 parsed->fhuuid = fsid;
 434                 break;
 435         case FSID_UUID16: /* 16 byte uuid */
 436                 if (fsidlen != 16)
 437                         return -1;
 438                 parsed->uuidlen = 16;
 439                 parsed->fhuuid = fsid;
 440                 break;
 441         case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
 442                 if (fsidlen != 24)
 443                         return -1;
 444                 memcpy(&inode64, fsid, 8);
 445                 parsed->inode = inode64;
 446                 parsed->uuidlen = 16;
 447                 parsed->fhuuid = fsid+8;
 448                 break;
 449         }
 450         return 0;
 451 }
 452
 453 static bool match_fsid(struct parsed_fsid *parsed, nfs_export *exp, char *path)
 454 {
 455         struct stat stb;
 456         int type;
 457         char u[16];
 458
 459         if (stat(path, &stb) != 0)
 460                 return false;
 461         if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode))
 462                 return false;
 463
 464         switch (parsed->fsidtype) {
 465         case FSID_DEV:
 466         case FSID_MAJOR_MINOR:
 467         case FSID_ENCODE_DEV:
 468                 if (stb.st_ino != parsed->inode)
 469                         return false;
 470                 if (parsed->major != major(stb.st_dev) ||
 471                     parsed->minor != minor(stb.st_dev))
 472                         return false;
 473                 return true;
 474         case FSID_NUM:
 475                 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
 476                      exp->m_export.e_fsid != parsed->fsidnum))
 477                         return false;
 478                 return true;
 479         case FSID_UUID4_INUM:
 480         case FSID_UUID16_INUM:
 481                 if (stb.st_ino != parsed->inode)
 482                         return false;
 483                 goto check_uuid;
 484         case FSID_UUID8:
 485         case FSID_UUID16:
 486                 if (!is_mountpoint(path))
 487                         return false;
 488         check_uuid:
 489                 if (exp->m_export.e_uuid)
 490                         get_uuid(exp->m_export.e_uuid, parsed->uuidlen, u);
 491                 else
 492                         for (type = 0;
 493                              uuid_by_path(path, type, parsed->uuidlen, u);
 494                              type++)
 495                                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) == 0)
 496                                         return true;
 497
 498                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) != 0)
 499                         return false;
 500                 return true;
 501         }
 502         /* Well, unreachable, actually: */
 503         return false;
 504 }
 505
 506 static struct addrinfo *lookup_client_addr(char *dom)
 507 {
 508         struct addrinfo *ret;
 509         struct addrinfo *tmp;
 510
 511         dom++; /* skip initial "$" */
 512
 513         tmp = host_pton(dom);
 514         if (tmp == NULL)
 515                 return NULL;
 516         ret = client_resolve(tmp->ai_addr);
 517         freeaddrinfo(tmp);
 518         return ret;
 519 }
 520
 521 static void nfsd_fh(FILE *f)
 522 {
 523         /* request are:
 524          *  domain fsidtype fsid
 525          * interpret fsid, find export point and options, and write:
 526          *  domain fsidtype fsid expiry path
 527          */
 528         char *cp;
 529         char *dom;
 530         int fsidtype;
 531         int fsidlen;
 532         char fsid[32];
 533         struct parsed_fsid parsed;
 534         struct exportent *found = NULL;
 535         struct addrinfo *ai = NULL;
 536         char *found_path = NULL;
 537         nfs_export *exp;
 538         int i;
 539         int dev_missing = 0;
 540
 541         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 542                 return;
 543
 544         xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
 545
 546         cp = lbuf;
 547
 548         dom = malloc(strlen(cp));
 549         if (dom == NULL)
 550                 return;
 551         if (qword_get(&cp, dom, strlen(cp)) <= 0)
 552                 goto out;
 553         if (qword_get_int(&cp, &fsidtype) != 0)
 554                 goto out;
 555         if (fsidtype < 0 || fsidtype > 7)
 556                 goto out; /* unknown type */
 557         if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
 558                 goto out;
 559         if (parse_fsid(fsidtype, fsidlen, fsid, &parsed))
 560                 goto out;
 561
 562         auth_reload();
 563
 564         if (is_ipaddr_client(dom)) {
 565                 ai = lookup_client_addr(dom);
 566                 if (!ai)
 567                         goto out;
 568         }
 569
 570         /* Now determine export point for this fsid/domain */
 571         for (i=0 ; i < MCL_MAXTYPES; i++) {
 572                 nfs_export *next_exp;
 573                 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
 574                         char *path;
 575
 576                         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
 577                                 static nfs_export *prev = NULL;
 578                                 static void *mnt = NULL;
 579
 580                                 if (prev == exp) {
 581                                         /* try a submount */
 582                                         path = next_mnt(&mnt, exp->m_export.e_path);
 583                                         if (!path) {
 584                                                 next_exp = exp->m_next;
 585                                                 prev = NULL;
 586                                                 continue;
 587                                         }
 588                                         next_exp = exp;
 589                                 } else {
 590                                         prev = exp;
 591                                         mnt = NULL;
 592                                         path = exp->m_export.e_path;
 593                                         next_exp = exp;
 594                                 }
 595                         } else {
 596                                 path = exp->m_export.e_path;
 597                                 next_exp = exp->m_next;
 598                         }
 599
 600                         if (!is_ipaddr_client(dom)
 601                                         && !namelist_client_matches(exp, dom))
 602                                 continue;
 603                         if (exp->m_export.e_mountpoint &&
 604                             !is_mountpoint(exp->m_export.e_mountpoint[0]?
 605                                            exp->m_export.e_mountpoint:
 606                                            exp->m_export.e_path))
 607                                 dev_missing ++;
 608
 609                         if (!match_fsid(&parsed, exp, path))
 610                                 continue;
 611                         if (is_ipaddr_client(dom)
 612                                         && !ipaddr_client_matches(exp, ai))
 613                                 continue;
 614                         if (!found || subexport(&exp->m_export, found)) {
 615                                 found = &exp->m_export;
 616                                 free(found_path);
 617                                 found_path = strdup(path);
 618                                 if (found_path == NULL)
 619                                         goto out;
 620                         } else if (strcmp(found->e_path, exp->m_export.e_path) != 0
 621                                    && !subexport(found, &exp->m_export))
 622                         {
 623                                 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
 624                                      found_path, path, dom);
 625                         } else {
 626                                 /* same path, if one is V4ROOT, choose the other */
 627                                 if (found->e_flags & NFSEXP_V4ROOT) {
 628                                         found = &exp->m_export;
 629                                         free(found_path);
 630                                         found_path = strdup(path);
 631                                         if (found_path == NULL)
 632                                                 goto out;
 633                                 }
 634                         }
 635                 }
 636         }
 637         if (found &&
 638             found->e_mountpoint &&
 639             !is_mountpoint(found->e_mountpoint[0]?
 640                            found->e_mountpoint:
 641                            found->e_path)) {
 642                 /* Cannot export this yet
 643                  * should log a warning, but need to rate limit
 644                    xlog(L_WARNING, "%s not exported as %d not a mountpoint",
 645                    found->e_path, found->e_mountpoint);
 646                  */
 647                 /* FIXME we need to make sure we re-visit this later */
 648                 goto out;
 649         }
 650         if (!found && dev_missing) {
 651                 /* The missing dev could be what we want, so just be
 652                  * quite rather than returning stale yet
 653                  */
 654                 goto out;
 655         }
 656
 657         if (found)
 658                 if (cache_export_ent(dom, found, found_path) < 0)
 659                         found = 0;
 660
 661         qword_print(f, dom);
 662         qword_printint(f, fsidtype);
 663         qword_printhex(f, fsid, fsidlen);
 664         /* The fsid -> path lookup can be quite expensive as it
 665          * potentially stats and reads lots of devices, and some of those
 666          * might have spun-down.  The Answer is not likely to
 667          * change underneath us, and an 'exportfs -f' can always
 668          * remove this from the kernel, so use a really log
 669          * timeout.  Maybe this should be configurable on the command
 670          * line.
 671          */
 672         qword_printint(f, 0x7fffffff);
 673         if (found)
 674                 qword_print(f, found_path);
 675         qword_eol(f);
 676  out:
 677         if (found_path)
 678                 free(found_path);
 679         freeaddrinfo(ai);
 680         free(dom);
 681         xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
 682         return;
 683 }
 684
 685 static void write_fsloc(FILE *f, struct exportent *ep)
 686 {
 687         struct servers *servers;
 688
 689         if (ep->e_fslocmethod == FSLOC_NONE)
 690                 return;
 691
 692         servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
 693         if (!servers)
 694                 return;
 695         qword_print(f, "fsloc");
 696         qword_printint(f, servers->h_num);
 697         if (servers->h_num >= 0) {
 698                 int i;
 699                 for (i=0; i<servers->h_num; i++) {
 700                         qword_print(f, servers->h_mp[i]->h_host);
 701                         qword_print(f, servers->h_mp[i]->h_path);
 702                 }
 703         }
 704         qword_printint(f, servers->h_referral);
 705         release_replicas(servers);
 706 }
 707
 708 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
 709 {
 710         struct sec_entry *p;
 711
 712         for (p = ep->e_secinfo; p->flav; p++)
 713                 ; /* Do nothing */
 714         if (p == ep->e_secinfo) {
 715                 /* There was no sec= option */
 716                 return;
 717         }
 718         qword_print(f, "secinfo");
 719         qword_printint(f, p - ep->e_secinfo);
 720         for (p = ep->e_secinfo; p->flav; p++) {
 721                 qword_printint(f, p->flav->fnum);
 722                 qword_printint(f, p->flags & flag_mask);
 723         }
 724
 725 }
 726
 727 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
 728 {
 729         qword_print(f, domain);
 730         qword_print(f, path);
 731         if (exp) {
 732                 int different_fs = strcmp(path, exp->e_path) != 0;
 733                 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
 734
 735                 qword_printtimefrom(f, exp->e_ttl);
 736                 qword_printint(f, exp->e_flags & flag_mask);
 737                 qword_printint(f, exp->e_anonuid);
 738                 qword_printint(f, exp->e_anongid);
 739                 qword_printint(f, exp->e_fsid);
 740                 write_fsloc(f, exp);
 741                 write_secinfo(f, exp, flag_mask);
 742                 if (exp->e_uuid == NULL || different_fs) {
 743                         char u[16];
 744                         if (uuid_by_path(path, 0, 16, u)) {
 745                                 qword_print(f, "uuid");
 746                                 qword_printhex(f, u, 16);
 747                         }
 748                 } else {
 749                         char u[16];
 750                         get_uuid(exp->e_uuid, 16, u);
 751                         qword_print(f, "uuid");
 752                         qword_printhex(f, u, 16);
 753                 }
 754         } else
 755                 qword_printtimefrom(f, DEFAULT_TTL);
 756         return qword_eol(f);
 757 }
 758
 759 static int is_subdirectory(char *child, char *parent)
 760 {
 761         size_t l = strlen(parent);
 762
 763         return strcmp(child, parent) == 0
 764                 || (strncmp(child, parent, l) == 0 && child[l] == '/');
 765 }
 766
 767 static int path_matches(nfs_export *exp, char *path)
 768 {
 769         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
 770                 return is_subdirectory(path, exp->m_export.e_path);
 771         return strcmp(path, exp->m_export.e_path) == 0;
 772 }
 773
 774 static int
 775 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
 776 {
 777         return path_matches(exp, path) && client_matches(exp, dom, ai);
 778 }
 779
 780 static nfs_export *
 781 lookup_export(char *dom, char *path, struct addrinfo *ai)
 782 {
 783         nfs_export *exp;
 784         nfs_export *found = NULL;
 785         int found_type = 0;
 786         int i;
 787
 788         for (i=0 ; i < MCL_MAXTYPES; i++) {
 789                 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
 790                         if (!export_matches(exp, dom, path, ai))
 791                                 continue;
 792                         if (!found) {
 793                                 found = exp;
 794                                 found_type = i;
 795                                 continue;
 796                         }
 797                         /* Always prefer non-V4ROOT exports */
 798                         if (exp->m_export.e_flags & NFSEXP_V4ROOT)
 799                                 continue;
 800                         if (found->m_export.e_flags & NFSEXP_V4ROOT) {
 801                                 found = exp;
 802                                 found_type = i;
 803                                 continue;
 804                         }
 805
 806                         /* If one is a CROSSMOUNT, then prefer the longest path */
 807                         if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
 808                              (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
 809                             strlen(found->m_export.e_path) !=
 810                             strlen(exp->m_export.e_path)) {
 811
 812                                 if (strlen(exp->m_export.e_path) >
 813                                     strlen(found->m_export.e_path)) {
 814                                         found = exp;
 815                                         found_type = i;
 816                                 }
 817                                 continue;
 818
 819                         } else if (found_type == i && found->m_warned == 0) {
 820                                 xlog(L_WARNING, "%s exported to both %s and %s, "
 821                                      "arbitrarily choosing options from first",
 822                                      path, found->m_client->m_hostname, exp->m_client->m_hostname,
 823                                      dom);
 824                                 found->m_warned = 1;
 825                         }
 826                 }
 827         }
 828         return found;
 829 }
 830
 831 #ifdef HAVE_NFS_PLUGIN_H
 832 #include <dlfcn.h>
 833 #include <nfs-plugin.h>
 834
 835 /*
 836  * Find the export entry for the parent of "pathname".
 837  * Caller must not free returned exportent.
 838  */
 839 static struct exportent *lookup_parent_export(char *dom,
 840                 const char *pathname, struct addrinfo *ai)
 841 {
 842         char *parent, *slash;
 843         nfs_export *result;
 844
 845         parent = strdup(pathname);
 846         if (parent == NULL) {
 847                 xlog(D_GENERAL, "%s: failed to allocate parent path buffer",
 848                         __func__);
 849                 goto out_default;
 850         }
 851         xlog(D_CALL, "%s: pathname = '%s'", __func__, pathname);
 852
 853 again:
 854         /* shorten pathname by one component */
 855         slash = strrchr(parent, '/');
 856         if (slash == NULL) {
 857                 xlog(D_GENERAL, "%s: no slash found in pathname",
 858                         __func__);
 859                 goto out_default;
 860         }
 861         *slash = '\0';
 862
 863         if (strlen(parent) == 0) {
 864                 result = lookup_export(dom, "/", ai);
 865                 if (result == NULL) {
 866                         xlog(L_ERROR, "%s: no root export found.", __func__);
 867                         goto out_default;
 868                 }
 869                 goto out;
 870         }
 871
 872         result = lookup_export(dom, parent, ai);
 873         if (result == NULL) {
 874                 xlog(D_GENERAL, "%s: lookup_export(%s) found nothing",
 875                         __func__, parent);
 876                 goto again;
 877         }
 878
 879 out:
 880         xlog(D_CALL, "%s: found export for %s", __func__, parent);
 881         free(parent);
 882         return &result->m_export;
 883
 884 out_default:
 885         free(parent);
 886         return mkexportent("*", "/", "insecure");
 887 }
 888
 889 /*
 890  * Walk through a set of FS locations and build an e_fslocdata string.
 891  * Returns true if all went to plan; otherwise, false.
 892  */
 893 static bool locations_to_fslocdata(struct jp_ops *ops,
 894                 nfs_fsloc_set_t locations, char *fslocdata,
 895                 size_t remaining, int *ttl)
 896 {
 897         char *server, *last_path, *rootpath, *ptr;
 898         _Bool seen = false;
 899
 900         last_path = NULL;
 901         rootpath = NULL;
 902         server = NULL;
 903         ptr = fslocdata;
 904         *ttl = 0;
 905
 906         for (;;) {
 907                 enum jp_status status;
 908                 int len;
 909
 910                 status = ops->jp_get_next_location(locations, &server,
 911                                                         &rootpath, ttl);
 912                 if (status == JP_EMPTY)
 913                         break;
 914                 if (status != JP_OK) {
 915                         xlog(D_GENERAL, "%s: failed to parse location: %s",
 916                                 __func__, ops->jp_error(status));
 917                         goto out_false;
 918                 }
 919                 xlog(D_GENERAL, "%s: Location: %s:%s",
 920                         __func__, server, rootpath);
 921
 922                 if (last_path && strcmp(rootpath, last_path) == 0) {
 923                         len = snprintf(ptr, remaining, "+%s", server);
 924                         if (len < 0) {
 925                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
 926                                 goto out_false;
 927                         }
 928                         if ((size_t)len >= remaining) {
 929                                 xlog(D_GENERAL, "%s: fslocdata buffer overflow", __func__);
 930                                 goto out_false;
 931                         }
 932                         remaining -= (size_t)len;
 933                         ptr += len;
 934                 } else {
 935                         if (last_path == NULL)
 936                                 len = snprintf(ptr, remaining, "%s@%s",
 937                                                         rootpath, server);
 938                         else
 939                                 len = snprintf(ptr, remaining, ":%s@%s",
 940                                                         rootpath, server);
 941                         if (len < 0) {
 942                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
 943                                 goto out_false;
 944                         }
 945                         if ((size_t)len >= remaining) {
 946                                 xlog(D_GENERAL, "%s: fslocdata buffer overflow",
 947                                         __func__);
 948                                 goto out_false;
 949                         }
 950                         remaining -= (size_t)len;
 951                         ptr += len;
 952                         last_path = rootpath;
 953                 }
 954
 955                 seen = true;
 956                 free(rootpath);
 957                 free(server);
 958         }
 959
 960         xlog(D_CALL, "%s: fslocdata='%s', ttl=%d",
 961                 __func__, fslocdata, *ttl);
 962         return seen;
 963
 964 out_false:
 965         free(rootpath);
 966         free(server);
 967         return false;
 968 }
 969
 970 /*
 971  * Duplicate the junction's parent's export options and graft in
 972  * the fslocdata we constructed from the locations list.
 973  */
 974 static struct exportent *create_junction_exportent(struct exportent *parent,
 975                 const char *junction, const char *fslocdata, int ttl)
 976 {
 977         static struct exportent *eep;
 978
 979         eep = (struct exportent *)malloc(sizeof(*eep));
 980         if (eep == NULL)
 981                 goto out_nomem;
 982
 983         dupexportent(eep, parent);
 984         strcpy(eep->e_path, junction);
 985         eep->e_hostname = strdup(parent->e_hostname);
 986         if (eep->e_hostname == NULL) {
 987                 free(eep);
 988                 goto out_nomem;
 989         }
 990         free(eep->e_uuid);
 991         eep->e_uuid = NULL;
 992         eep->e_ttl = (unsigned int)ttl;
 993
 994         free(eep->e_fslocdata);
 995         eep->e_fslocdata = strdup(fslocdata);
 996         if (eep->e_fslocdata == NULL) {
 997                 free(eep->e_hostname);
 998                 free(eep);
 999                 goto out_nomem;
1000         }
1001         eep->e_fslocmethod = FSLOC_REFER;
1002         return eep;
1003
1004 out_nomem:
1005         xlog(L_ERROR, "%s: No memory", __func__);
1006         return NULL;
1007 }
1008
1009 /*
1010  * Walk through the set of FS locations and build an exportent.
1011  * Returns pointer to an exportent if "junction" refers to a junction.
1012  */
1013 static struct exportent *locations_to_export(struct jp_ops *ops,
1014                 nfs_fsloc_set_t locations, const char *junction,
1015                 struct exportent *parent)
1016 {
1017         static char fslocdata[BUFSIZ];
1018         int ttl;
1019
1020         fslocdata[0] = '\0';
1021         if (!locations_to_fslocdata(ops, locations,
1022                                         fslocdata, sizeof(fslocdata), &ttl))
1023                 return NULL;
1024         return create_junction_exportent(parent, junction, fslocdata, ttl);
1025 }
1026
1027 /*
1028  * Retrieve locations information in "junction" and dump it to the
1029  * kernel.  Returns pointer to an exportent if "junction" refers
1030  * to a junction.
1031  */
1032 static struct exportent *invoke_junction_ops(void *handle, char *dom,
1033                 const char *junction, struct addrinfo *ai)
1034 {
1035         struct exportent *parent, *exp = NULL;
1036         nfs_fsloc_set_t locations;
1037         enum jp_status status;
1038         struct jp_ops *ops;
1039         char *error;
1040
1041         ops = (struct jp_ops *)dlsym(handle, "nfs_junction_ops");
1042         error = dlerror();
1043         if (error != NULL) {
1044                 xlog(D_GENERAL, "%s: dlsym(jp_junction_ops): %s",
1045                         __func__, error);
1046                 return NULL;
1047         }
1048         if (ops->jp_api_version != JP_API_VERSION) {
1049                 xlog(D_GENERAL, "%s: unrecognized junction API version: %u",
1050                         __func__, ops->jp_api_version);
1051                 return NULL;
1052         }
1053
1054         status = ops->jp_init(false);
1055         if (status != JP_OK) {
1056                 xlog(D_GENERAL, "%s: failed to resolve %s: %s",
1057                         __func__, junction, ops->jp_error(status));
1058                 return NULL;
1059         }
1060
1061         status = ops->jp_get_locations(junction, &locations);
1062         switch (status) {
1063         case JP_OK:
1064                 break;
1065         case JP_NOTJUNCTION:
1066                 xlog(D_GENERAL, "%s: %s is not a junction",
1067                         __func__, junction);
1068                 goto out;
1069         default:
1070                 xlog(L_WARNING, "Dangling junction %s: %s",
1071                         junction, ops->jp_error(status));
1072                 goto out;
1073         }
1074
1075         parent = lookup_parent_export(dom, junction, ai);
1076         if (parent == NULL)
1077                 goto out;
1078
1079         exp = locations_to_export(ops, locations, junction, parent);
1080
1081         ops->jp_put_locations(locations);
1082
1083 out:
1084         ops->jp_done();
1085         return exp;
1086 }
1087
1088 /*
1089  * Load the junction plug-in, then try to resolve "pathname".
1090  * Returns pointer to an initialized exportent if "junction"
1091  * refers to a junction, or NULL if not.
1092  */
1093 static struct exportent *lookup_junction(char *dom, const char *pathname,
1094                 struct addrinfo *ai)
1095 {
1096         struct exportent *exp;
1097         void *handle;
1098
1099         handle = dlopen("libnfsjunct.so", RTLD_NOW);
1100         if (handle == NULL) {
1101                 xlog(D_GENERAL, "%s: dlopen: %s", __func__, dlerror());
1102                 return NULL;
1103         }
1104         (void)dlerror();        /* Clear any error */
1105
1106         exp = invoke_junction_ops(handle, dom, pathname, ai);
1107
1108         /* We could leave it loaded to make junction resolution
1109          * faster next time.  However, if we want to replace the
1110          * library, that would require restarting mountd. */
1111         (void)dlclose(handle);
1112         return exp;
1113 }
1114
1115 static void lookup_nonexport(FILE *f, char *dom, char *path,
1116                 struct addrinfo *ai)
1117 {
1118         struct exportent *eep;
1119
1120         eep = lookup_junction(dom, path, ai);
1121         dump_to_cache(f, dom, path, eep);
1122         if (eep == NULL)
1123                 return;
1124         exportent_release(eep);
1125         free(eep);
1126 }
1127 #else   /* !HAVE_NFS_PLUGIN_H */
1128 static void lookup_nonexport(FILE *f, char *dom, char *path,
1129                 struct addrinfo *UNUSED(ai))
1130 {
1131         dump_to_cache(f, dom, path, NULL);
1132 }
1133 #endif  /* !HAVE_NFS_PLUGIN_H */
1134
1135 static void nfsd_export(FILE *f)
1136 {
1137         /* requests are:
1138          *  domain path
1139          * determine export options and return:
1140          *  domain path expiry flags anonuid anongid fsid
1141          */
1142
1143         char *cp;
1144         char *dom, *path;
1145         nfs_export *found = NULL;
1146         struct addrinfo *ai = NULL;
1147
1148         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
1149                 return;
1150
1151         xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
1152
1153         cp = lbuf;
1154         dom = malloc(strlen(cp));
1155         path = malloc(strlen(cp));
1156
1157         if (!dom || !path)
1158                 goto out;
1159
1160         if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
1161                 goto out;
1162         if (qword_get(&cp, path, strlen(lbuf)) <= 0)
1163                 goto out;
1164
1165         auth_reload();
1166
1167         if (is_ipaddr_client(dom)) {
1168                 ai = lookup_client_addr(dom);
1169                 if (!ai)
1170                         goto out;
1171         }
1172
1173         found = lookup_export(dom, path, ai);
1174
1175         if (found) {
1176                 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
1177                         xlog(L_WARNING,
1178                              "Cannot export %s, possibly unsupported filesystem"
1179                              " or fsid= required", path);
1180                         dump_to_cache(f, dom, path, NULL);
1181                 }
1182         } else
1183                 lookup_nonexport(f, dom, path, ai);
1184
1185  out:
1186         xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
1187         if (dom) free(dom);
1188         if (path) free(path);
1189         freeaddrinfo(ai);
1190 }
1191
1192
1193 struct {
1194         char *cache_name;
1195         void (*cache_handle)(FILE *f);
1196         FILE *f;
1197         char vbuf[RPC_CHAN_BUF_SIZE];
1198 } cachelist[] = {
1199         { "auth.unix.ip", auth_unix_ip, NULL, ""},
1200         { "auth.unix.gid", auth_unix_gid, NULL, ""},
1201         { "nfsd.export", nfsd_export, NULL, ""},
1202         { "nfsd.fh", nfsd_fh, NULL, ""},
1203         { NULL, NULL, NULL, ""}
1204 };
1205
1206 extern int manage_gids;
1207
1208 /**
1209  * cache_open - prepare communications channels with kernel RPC caches
1210  *
1211  */
1212 void cache_open(void)
1213 {
1214         int i;
1215         for (i=0; cachelist[i].cache_name; i++ ) {
1216                 char path[100];
1217                 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
1218                         continue;
1219                 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
1220                 cachelist[i].f = fopen(path, "r+");
1221                 if (cachelist[i].f != NULL) {
1222                         setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF,
1223                                 RPC_CHAN_BUF_SIZE);
1224                 }
1225         }
1226 }
1227
1228 /**
1229  * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
1230  * @fdset: pointer to fd_set to prepare
1231  */
1232 void cache_set_fds(fd_set *fdset)
1233 {
1234         int i;
1235         for (i=0; cachelist[i].cache_name; i++) {
1236                 if (cachelist[i].f)
1237                         FD_SET(fileno(cachelist[i].f), fdset);
1238         }
1239 }
1240
1241 /**
1242  * cache_process_req - process any active cache file descriptors during service loop iteration
1243  * @fdset: pointer to fd_set to examine for activity
1244  */
1245 int cache_process_req(fd_set *readfds)
1246 {
1247         int i;
1248         int cnt = 0;
1249         for (i=0; cachelist[i].cache_name; i++) {
1250                 if (cachelist[i].f != NULL &&
1251                     FD_ISSET(fileno(cachelist[i].f), readfds)) {
1252                         cnt++;
1253                         cachelist[i].cache_handle(cachelist[i].f);
1254                         FD_CLR(fileno(cachelist[i].f), readfds);
1255                 }
1256         }
1257         return cnt;
1258 }
1259
1260
1261 /*
1262  * Give IP->domain and domain+path->options to kernel
1263  * % echo nfsd $IP  $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
1264  * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
1265  */
1266
1267 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
1268 {
1269         int err;
1270         FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
1271         if (!f)
1272                 return -1;
1273
1274         err = dump_to_cache(f, domain, exp->e_path, exp);
1275         if (err) {
1276                 xlog(L_WARNING,
1277                      "Cannot export %s, possibly unsupported filesystem or"
1278                      " fsid= required", exp->e_path);
1279         }
1280
1281         while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
1282                 /* really an 'if', but we can break out of
1283                  * a 'while' more easily */
1284                 /* Look along 'path' for other filesystems
1285                  * and export them with the same options
1286                  */
1287                 struct stat stb;
1288                 size_t l = strlen(exp->e_path);
1289                 __dev_t dev;
1290
1291                 if (strlen(path) <= l || path[l] != '/' ||
1292                     strncmp(exp->e_path, path, l) != 0)
1293                         break;
1294                 if (stat(exp->e_path, &stb) != 0)
1295                         break;
1296                 dev = stb.st_dev;
1297                 while(path[l] == '/') {
1298                         char c;
1299                         /* errors for submount should fail whole filesystem */
1300                         int err2;
1301
1302                         l++;
1303                         while (path[l] != '/' && path[l])
1304                                 l++;
1305                         c = path[l];
1306                         path[l] = 0;
1307                         err2 = lstat(path, &stb);
1308                         path[l] = c;
1309                         if (err2 < 0)
1310                                 break;
1311                         if (stb.st_dev == dev)
1312                                 continue;
1313                         dev = stb.st_dev;
1314                         path[l] = 0;
1315                         dump_to_cache(f, domain, path, exp);
1316                         path[l] = c;
1317                 }
1318                 break;
1319         }
1320
1321         fclose(f);
1322         return err;
1323 }
1324
1325 /**
1326  * cache_export - Inform kernel of a new nfs_export
1327  * @exp: target nfs_export
1328  * @path: NUL-terminated C string containing export path
1329  */
1330 int cache_export(nfs_export *exp, char *path)
1331 {
1332         char buf[INET6_ADDRSTRLEN];
1333         int err;
1334         FILE *f;
1335
1336         f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
1337         if (!f)
1338                 return -1;
1339
1340
1341         qword_print(f, "nfsd");
1342         qword_print(f,
1343                 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
1344         qword_printtimefrom(f, exp->m_export.e_ttl);
1345         qword_print(f, exp->m_client->m_hostname);
1346         err = qword_eol(f);
1347
1348         fclose(f);
1349
1350         err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
1351                 || err;
1352         return err;
1353 }
1354
1355 /**
1356  * cache_get_filehandle - given an nfs_export, get its root filehandle
1357  * @exp: target nfs_export
1358  * @len: length of requested file handle
1359  * @p: NUL-terminated C string containing export path
1360  *
1361  * Returns pointer to NFS file handle of root directory of export
1362  *
1363  * {
1364  *   echo $domain $path $length
1365  *   read filehandle <&0
1366  * } <> /proc/fs/nfsd/filehandle
1367  */
1368 struct nfs_fh_len *
1369 cache_get_filehandle(nfs_export *exp, int len, char *p)
1370 {
1371         FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1372         char buf[200];
1373         char *bp = buf;
1374         int failed;
1375         static struct nfs_fh_len fh;
1376
1377         if (!f)
1378                 f = fopen("/proc/fs/nfs/filehandle", "r+");
1379         if (!f)
1380                 return NULL;
1381
1382         qword_print(f, exp->m_client->m_hostname);
1383         qword_print(f, p);
1384         qword_printint(f, len);
1385         failed = qword_eol(f);
1386
1387         if (!failed)
1388                 failed = (fgets(buf, sizeof(buf), f) == NULL);
1389         fclose(f);
1390         if (failed)
1391                 return NULL;
1392         memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1393         fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);
1394         return &fh;
1395 }